Georgios Pitsilis

Georgios K. Pitsilis

Hate speech is a form of online harassment that involves the use of abusive language, and it is commonly seen in social media posts. This sort of harassment mainly focuses on specific group characteristics such as religion, gender, ethnicity, etc and it has both societal and economic consequences nowadays. The automatic detection of abusive language in text postings has always been a difficult task, but it is lately receiving much interest from the scientific community. This paper addresses the important problem of discerning hateful content in social media. The model we propose in this work is an extension of an existing approach based on LSTM neural network architectures, which we appropriately enhanced and fine-tuned to detect certain forms of hatred language, such as racism or sexism, in a short text. The most significant enhancement is the conversion to a two-stage scheme consisting of Recurrent Neural Network (RNN) classifiers. The output of all One-vs-Rest (OvR) classifiers from the first stage are combined and used to train the second stage classifier, which finally determines the type of harassment. Our study includes a performance comparison of several proposed alternative methods for the second stage evaluated on a public corpus of 16k tweets, followed by a generalization study on another dataset. The reported results show the superior classification quality of the proposed scheme in the task of hate speech detection as compared to the current state-of-the-art.

Georgios K. Pitsilis, Heri Ramampiaro, Helge Langseth

This work addresses the challenges related to attacks on collaborative tagging systems, which often comes in a form of malicious annotations or profile injection attacks. In particular, we study various countermeasures against two types of such attacks for social tagging systems, the Overload attack and the Piggyback attack. The countermeasure schemes studied here include baseline classifiers such as, Naive Bayes filter and Support Vector Machine, as well as a Deep Learning approach. Our evaluation performed over synthetic spam data generated from del.icio.us dataset, shows that in most cases, Deep Learning can outperform the classical solutions, providing high-level protection against threats.

Georgios Pitsilis, Heri Ramampiaro, Helge Langseth

This work addresses challenges related to attacks on social tagging systems, which often comes in a form of malicious annotations or profile injection attacks. In particular, we study various countermeasures against two types of threats for such systems, the Overload and the Piggyback attacks. The studied countermeasures include baseline classifiers such as, Naive Bayes filter and Support Vector Machine, as well as a deep learning-based approach. Our evaluation performed over synthetic spam data, generated from del.icio.us, shows that in most cases, the deep learning-based approach provides the best protection against threats.

Georgios K. Pitsilis, Heri Ramampiaro, Helge Langseth

This paper addresses the important problem of discerning hateful content in social media. We propose a detection scheme that is an ensemble of Recurrent Neural Network (RNN) classifiers, and it incorporates various features associated with user-related information, such as the users' tendency towards racism or sexism. These data are fed as input to the above classifiers along with the word frequency vectors derived from the textual content. Our approach has been evaluated on a publicly available corpus of 16k tweets, and the results demonstrate its effectiveness in comparison to existing state of the art solutions. More specifically, our scheme can successfully distinguish racism and sexism messages from normal text, and achieve higher classification quality than current state-of-the-art algorithms.

Georgios K. Pitsilis, Mohamed El-Hadedy

Micro-blogs are contemporary broadcasting services for exchanging small elements of content, including video and images. Despite its popularity, micro-blogging is not without issues. So far, various security concerns, such as: privacy and confidentiality of micro-blogging systems have attracted the interest of the scientific community. Nevertheless, in this document we refer to a security issue that is concerned with the posting and circulation of fake messages. With the aim to make micro-blogs a credible source of information, in this work we propose a protocol, as a solution, that can be easily adapted to existing services such as: Twitter and Facebook. With the main keys of the solution being the employment of digital signatures and watermarks, the protocol can serve postings of various types of content. Finally, the solution serves the dual purpose of proofing the fake messages as well as the repudiation of postings.

Georgios Pitsilis, Wei Wang

Social bookmarking and tagging has emerged a new era in user collaboration. Collaborative Tagging allows users to annotate content of their liking, which via the appropriate algorithms can render useful for the provision of product recommendations. It is the case today for tag-based algorithms to work complementary to rating-based recommendation mechanisms to predict the user liking to various products. In this paper we propose an alternative algorithm for computing personalized recommendations of products, that uses exclusively the tags provided by the users. Our approach is based on the idea of using the semantic similarity of the user-provided tags for clustering them into groups of similar meaning. Afterwards, some measurable characteristics of users' Annotation Competency are combined with other metrics, such as user similarity, for computing predictions. The evaluation on data used from a real-world collaborative tagging system, citeUlike, confirmed that our approach outperforms the baseline Vector Space model, as well as other state of the art algorithms, predicting the user liking more accurately.

Mohamed El-Hadedy, Georgios Pitsilis, Svein J. Knapskog

Many electronic content providers today like Flickr and Google, offer space to users to publish their electronic media (e.g. photos and videos) in their cloud infrastructures, so that they can be publicly accessed. Features like including other information, such as keywords or owner information into the digital material is already offered by existing providers. Despite the useful features made available to users by such infrastructures, the authorship of the published content is not protected against various attacks such as compression. In this paper we propose a robust scheme that uses digital invisible watermarking and hashing to protect the authorship of the digital content and provide resistance against malicious manipulation of multimedia content. The scheme is enhanced by an algorithm called MMBEC, that is an extension of an established scheme MBEC, towards higher resistance.

Georgios Pitsilis, Svein J. Knapskog

Trust has been explored by many researchers in the past as a successful solution for assisting recommender systems. Even though the approach of using a web-of-trust scheme for assisting the recommendation production is well adopted, issues like the sparsity problem have not been explored adequately so far with regard to this. In this work we are proposing and testing a scheme that uses the existing ratings of users to calculate the hypothetical trust that might exist between them. The purpose is to demonstrate how some basic social networking when applied to an existing system can help in alleviating problems of traditional recommender system schemes. Interestingly, such schemes are also alleviating the cold start problem from which mainly new users are suffering. In order to show how good the system is in that respect, we measure the performance at various times as the system evolves and we also contrast the solution with existing approaches. Finally, we present the results which justify that such schemes undoubtedly work better than a system that makes no use of trust at all.